Method and device for realizing sensitive data identification under Xinchuang environment, processor and computer readable storage medium thereof

Abstract

The present application relates to a kind of sensitive data identification method for realizing under Xinxing environment, wherein, the method includes the following steps: collecting metadata and sample data;Data pre-processing and data set division are carried out to the data collected;Sensitivedata sample is constructed, and attention enhancement operation is carried out;Based on the target function of sensitivedata identification sample construction;The target function is used with deep neural network model, and the construction and training of model are carried out;Sensitivedata is identified using the network model of training completion and is handled.The present application also relates to a corresponding device, processor and computer readable storage medium thereof.The sensitive data identification method for realizing under Xinxing environment, device, processor and computer readable storage medium thereof using the present application, can comprehensively consider the concealment of sensitivedata in big data, the attention of specific characteristics is promoted, can effectively improve the robustness and generalization ability of model and improve the precision of sensitivedata identification.

Description

Technical Field

[0001] This invention relates to the field of computer software technology, and more particularly to the field of big data processing technology, specifically to a method, apparatus, processor, and computer-readable storage medium for identifying sensitive data in a domestically developed information technology environment. Background Technology

[0002] With the continuous development of the big data era, the data resources faced by various industries and enterprises have become more complex and are constantly changing. Traditional data governance measures and methods have become less applicable to the new information technology innovation environment, prompting innovation in data governance strategies. Especially in the context of the information technology innovation environment, data types have become more diversified, data sources are broader, and the data volume is increasing rapidly, placing higher demands on data security governance. Sensitive data flows through many channels, spanning the entire data lifecycle and covering multiple stages such as data generation, analysis, statistics, transfer, and invalidation. Sensitive data leakage is most likely to occur during the flow of data to less controlled environments, such as when transferring data from a production environment to a testing environment or analysis system. Therefore, identifying sensitive data throughout the entire data lifecycle and then obfuscating it is of paramount importance. Current technology relies on manual identification of sensitive data, with database administrators searching and confirming sensitive data based on their personal experience. This method is time-consuming and labor-intensive, and has a high probability of missing sensitive data. Therefore, there is an urgent need to research an efficient sensitive data identification method to automatically identify sensitive data and reduce the probability of missing sensitive data. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, apparatus, processor and computer-readable storage medium for sensitive data identification in the context of information technology innovation, which can identify sensitive data more efficiently and conveniently.

[0004] To achieve the above objectives, the present invention provides a method, apparatus, processor, and computer-readable storage medium for sensitive data identification in a domestically developed information technology environment, as follows:

[0005] The method for identifying sensitive data in this information technology application innovation environment is characterized by the following steps:

[0006] (1) Collect metadata and sample data;

[0007] (2) Perform data preprocessing and dataset partitioning on the collected data;

[0008] (3) Construct sensitive data samples and perform attention enhancement operations;

[0009] (4) Construct an objective function based on samples identified from sensitive data;

[0010] (5) The objective function is constructed and trained using a deep neural network model;

[0011] (6) Use the trained network model to identify and process sensitive data.

[0012] Preferably, step (1) specifically includes the following steps:

[0013] (1.1) Collect a specified number of sample data from the business data table by traversing the data table, thereby obtaining structured and unstructured data samples and metadata information in the database and file system;

[0014] (1.2) Filter and generalize the collected metadata and sample data to remove impurity data.

[0015] Preferably, step (2) specifically includes the following steps:

[0016] (2.1) For the character features in the collected relevant data, one-hot encoding is used for numerical conversion;

[0017] (2.2) The MinMaxScaler() function is used to normalize the collected data so that the preprocessed data is limited to the range [0,1].

[0018] (2.3) Divide the data into a training set and a test set, and label the training set. The training set is used to train the model, and the test set is used to verify the accuracy of the model.

[0019] Preferably, step (3) specifically includes the following steps:

[0020] (3.1) Construct a sensitive data identification sample dataset as X[X1,X2,X3,…X] n ], and reference the query vector q;

[0021] (3.2) Identify the association between the sample data set X and the query vector q based on the sensitive data, obtain a score result, and input the score result into the softmax function for normalization mapping to obtain the attention distribution of each sample;

[0022] (3.3) Finally, based on the calculated attention distribution results, attention selection is performed on the sensitive data identification sample data set X, and important information is assigned a higher weight, while unimportant information is assigned a lower weight.

[0023] (3.4) Repeat the iterative process based on step (3.3) to obtain the final attention-enhanced sample D.

[0024] Preferably, step (3.2) is specifically implemented through the following calculation formula:

[0025]

[0026] Where q is the query vector, q′ is the vector length, and x t Let t be the t-th feature of the input sample, t∈[1,2,3,…,T], where T is the number of features in the sample. The scoring function S(x,q) is calculated as follows:

[0027] b t =s(x t ,q)=a t T ×tanh(Wx t +Uq)

[0028] Among them, a t T For attention distribution a t The transpose of x t Let W be the t-th feature of the input sample, and U be the parameters that are learned and adjusted during training and are set manually. They are generally values ​​in the range [0,1].

[0029] Preferably, the attention-enhanced samples D{D1,D2,…,D n}, where D in D i Specifically, it is obtained through the following formula:

[0030] D i =b t ·x t

[0031] Where t∈[1,2,3,…,T].

[0032] Preferably, step (4) specifically includes the following steps:

[0033] (4.1) Suppose that the attention-enhanced sample D contains c categories to be identified, and the set of categories to which it belongs is C{C1,C2,…,C…} n The set of probability matrices for all its categories is P{P1,P2,…,P}. n};

[0034] (4.2) Construct two function mappings to obtain the word feature set F{F1,F2,…,F} corresponding to each text. n} and the sentence feature set F′={F′1,F′2,…,F′} corresponding to each text. n}, where F = f(D), F′ = f′(D);

[0035] (4.3) Based on the above steps, obtain the final set of each text feature F″={F′1′,F′2′,…,F′ n '},in,

[0036] F″=W k FW k 'F'

[0037] Among them, W k W is the weight matrix of word features. k The weight matrix is ​​constructed using sentence features.

[0038] (4.4) Construct a classification function based on the final feature set F″ of each text obtained, and transform each final text feature into a probability set corresponding to each category. The calculation formula is as follows:

[0039]

[0040] The softmax function is a normalized exponential function.

[0041] (4.5) Based on the probability set of each category and the probability matrix set of all categories to which it truly belongs, the error function loss is calculated in the following way:

[0042]

[0043] (4.6) By continuously adjusting various functions and parameters, the error is reduced, and the objective function is transformed into function T. m This is used as the loss function for subsequent deep neural network models, specifically:

[0044]

[0045] Preferably, step (5) specifically includes the following steps:

[0046] (5.1) A deep neural network model is used for model construction;

[0047] (5.2) The network parameters of the deep neural network model are set as follows: the number of neural network layers is 8, the number of neurons in the middle layer is 100, the learning rate is 1.5, the number of neurons in the last layer is c, the activation function is the Sigmoid function, and the loss function is T. m ;

[0048] (5.3) The loss function T is made easier by using the gradient descent algorithm. m Minimize, and thus obtain T m When minimized, the set of class probabilities

[0049] (5.4) The following method is used to obtain the class probability set. In the middle, obtain the maximum value:

[0050]

[0051] Currently, the P mentioned above max The corresponding category is the output category of the deep neural network model.

[0052] Preferably, step (6) specifically includes:

[0053] After the deep neural network model is trained using training samples, the trained deep neural network model is used to classify and identify unlabeled samples, and the output category of the deep neural network model is used to determine whether the input data is sensitive data, so as to complete the final data identification.

[0054] The device for identifying sensitive data in the context of information technology innovation is characterized by comprising:

[0055] A processor is configured to execute computer-executable instructions;

[0056] The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the various steps of the method for sensitive data identification under the aforementioned information technology innovation environment.

[0057] The processor used to identify sensitive data in the context of information technology innovation is characterized in that the processor is configured to execute computer-executable instructions, and when the computer-executable instructions are executed by the processor, they implement the various steps of the method for identifying sensitive data in the context of information technology innovation described above.

[0058] The main feature of this computer-readable storage medium is that it stores a computer program thereon, which can be executed by a processor to implement the various steps of the method for sensitive data identification in the context of information technology innovation described above.

[0059] The method, apparatus, processor, and computer-readable storage medium for sensitive data identification in the context of the present invention can effectively improve the accuracy of sensitive data identification. In particular, the present invention uses an attention mechanism and constructs a specific loss function, which can comprehensively consider the concealment of sensitive data in big data and enhance attention to specific features, thereby effectively improving the robustness and generalization ability of the model. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method for sensitive data identification in the context of the present invention. Detailed Implementation

[0061] To more clearly describe the technical content of the present invention, the following description is provided in conjunction with specific embodiments.

[0062] Before describing the embodiments of the present invention in detail, it should be noted that, in the following, the terms “comprising,” “including,” or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0063] Please see Figure 1 As shown, this method for identifying sensitive data in the context of information technology innovation includes the following steps:

[0064] (1) Collect metadata and sample data;

[0065] (2) Perform data preprocessing and dataset partitioning on the collected data;

[0066] (3) Construct sensitive data samples and perform attention enhancement operations;

[0067] (4) Construct an objective function based on samples identified from sensitive data;

[0068] (5) The objective function is constructed and trained using a deep neural network model;

[0069] (6) Use the trained network model to identify and process sensitive data.

[0070] In a preferred embodiment of the present invention, step (1) specifically includes the following steps:

[0071] (1.1) Collect a specified number of sample data from the business data table by traversing the data table, thereby obtaining structured and unstructured data samples and metadata information in the database and file system;

[0072] (1.2) Filter and generalize the collected metadata and sample data to remove impurity data.

[0073] In a preferred embodiment of the present invention, step (2) specifically includes the following steps:

[0074] (2.1) For the character features in the collected relevant data, one-hot encoding is used for numerical conversion;

[0075] (2.2) The MinMaxScaler() function is used to normalize the collected data so that the preprocessed data is limited to the range [0,1].

[0076] (2.3) Divide the data into a training set and a test set, and label the training set. The training set is used to train the model, and the test set is used to verify the accuracy of the model.

[0077] In a preferred embodiment of the present invention, step (3) specifically includes the following steps:

[0078] (3.1) Construct a sensitive data identification sample dataset as X[X1,X2,X3,…X] n ], and reference the query vector q;

[0079] (3.2) Identify the association between the sample data set X and the query vector q based on the sensitive data, obtain a score result, and input the score result into the softmax function for normalization mapping to obtain the attention distribution of each sample;

[0080] (3.3) Finally, based on the calculated attention distribution results, attention selection is performed on the sensitive data identification sample data set X, and important information is assigned a higher weight, while unimportant information is assigned a lower weight.

[0081] (3.4) Repeat the iterative process based on step (3.3) to obtain the final attention-enhanced sample D.

[0082] In a preferred embodiment of the present invention, step (3.2) is specifically implemented using the following calculation formula:

[0083]

[0084] Where q is the query vector, q′ is the vector length, and x t Let t be the t-th feature of the input sample, t∈[1,2,3,…,T], where T is the number of features in the sample. The scoring function S(x,q) is calculated as follows:

[0085] b t =s(x t ,q)=a t T ×tanh(Wx t +Uq)

[0086] Among them, a t T For attention distribution a t The transpose of x t Let W be the t-th feature of the input sample, and U be the parameters that are learned and adjusted during training and are set manually. They are generally values ​​in the range [0,1].

[0087] As a preferred embodiment of the present invention, the attention-enhanced sample D{D1,D2,…,D n}, where D in D i Specifically, it is obtained through the following formula:

[0088] D i =b t ·x t

[0089] Where t∈[1,2,3,…,T].

[0090] In a preferred embodiment of the present invention, step (4) specifically includes the following steps:

[0091] (4.1) Suppose that the attention-enhanced sample D contains c categories to be identified, and the set of categories to which it belongs is C{C1,C2,…,C…} n The set of probability matrices for all its categories is P{P1,P2,…,P}. n};

[0092] (4.2) Construct two function mappings to obtain the word feature set F{F1,F2,…,F} corresponding to each text. n} and the sentence feature set F′={F′1,F′2,…,F′} corresponding to each text. n}, where F = f(D), F′ = f′(D);

[0093] (4.3) Based on the above steps, obtain the final set of each text feature F″={F′1′,F′2′,…,F′ n '},in,

[0094] F″=W k F+W k 'F'

[0095] Among them, Wk W is the weight matrix of word features. k The weight matrix is ​​constructed using sentence features.

[0096] (4.4) Construct a classification function based on the final feature set F″ of each text obtained, and transform each final text feature into a probability set corresponding to each category. The calculation formula is as follows:

[0097]

[0098] The softmax function is a normalized exponential function.

[0099] (4.5) Based on the probability set of each category and the probability matrix set of all categories to which it truly belongs, the error function loss is calculated in the following way:

[0100]

[0101] (4.6) By continuously adjusting various functions and parameters, the error is reduced, and the objective function is transformed into function T. m This is used as the loss function for subsequent deep neural network models, specifically:

[0102]

[0103] In a preferred embodiment of the present invention, step (5) specifically includes the following steps:

[0104] (5.1) A deep neural network model is used for model construction;

[0105] (5.2) The network parameters of the deep neural network model are set as follows: the number of neural network layers is 8, the number of neurons in the middle layer is 100, the learning rate is 1.5, the number of neurons in the last layer is c, the activation function is the Sigmoid function, and the loss function is T. m ;

[0106] (5.3) The loss function T is made easier by using the gradient descent algorithm. m Minimize, and thus obtain T m When minimized, the set of class probabilities

[0107] (5.4) The following method is used to obtain the class probability set. In the middle, obtain the maximum value:

[0108]

[0109] Currently, the P mentioned above maxThe corresponding category is the output category of the deep neural network model.

[0110] In a preferred embodiment of the present invention, step (6) specifically comprises:

[0111] After the deep neural network model is trained using training samples, the trained deep neural network model is used to classify and identify unlabeled samples, and the output category of the deep neural network model is used to determine whether the input data is sensitive data, so as to complete the final data identification.

[0112] The device for identifying sensitive data in a domestically developed information technology environment includes:

[0113] A processor is configured to execute computer-executable instructions;

[0114] The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the various steps of the method for sensitive data identification under the aforementioned information technology innovation environment.

[0115] The processor for implementing sensitive data identification in the context of information technology innovation is configured to execute computer-executable instructions, which, when executed by the processor, implement the various steps of the method for implementing sensitive data identification in the context of information technology innovation described above.

[0116] The main feature of this computer-readable storage medium is that it stores a computer program thereon, which can be executed by a processor to implement the various steps of the method for sensitive data identification in the context of information technology innovation described above.

[0117] In practical applications, the main steps of this technical solution for sensitive data identification in the context of domestic IT innovation are as follows:

[0118] Step 1: Collect metadata and sample data

[0119] In an enterprise's business system, a specified number of sample data are collected from business data tables by simultaneously traversing the data tables, obtaining structured and unstructured data samples and metadata information from the database and file system. Then, this data is filtered and generalized to remove "impurities" and improve the quality of the data samples, thereby reducing the computational load in the sensitive data identification and classification process.

[0120] Step 2: Data Preprocessing and Dataset Partitioning

[0121] For character-type features, one-hot encoding is used for numerical conversion. To avoid the influence of different dimensions between data flow features, the MinMaxScaler() function is used to normalize the data, limiting the preprocessed data to the range [0,1]. Then, the data is divided into training and test sets, and the training set is labeled. The training set is used for model training, and the test set is used to verify the model's accuracy.

[0122] Step 3: Perform attention enhancement operations on the samples.

[0123] Suppose the constructed sensitive data identification sample dataset is X[X1,X2,X3,…X]. n The attention mechanism is applied to the model input X for computation, using a query vector q. A scoring function is then used to calculate the correlation between the model input X and the query vector q, outputting a score. This score is then fed into a softmax function for normalization mapping, resulting in the attention distribution a[a1, a2, a3, ... a] corresponding to the query vector q on the model input X. n The formula for calculating attention distribution is as follows:

[0124]

[0125] Where q is the query vector, q′ is the vector length, and x t Let t be the t-th feature of the input sample, t∈[1,2,3,…,T], where T is the number of features in the sample. The scoring function S(x,q) is calculated using the following method:

[0126] b t =s(x t ,q)=a t T ×tanh(Wx t +Uq)

[0127] Among them, a t T It is the attention distribution a t The transpose of x t It is the t-th feature of the input sample. W and U are parameters that can be learned and adjusted during training. They are set manually and generally take values ​​in the range [0,1].

[0128] Ultimately, attention can be selected from the model's input using an attention distribution, assigning higher weights to important information and lower weights to less important information. This process is iterated continuously to obtain attention-enhanced samples D. D{D1,D2,…,D… n}, D in set D i The formula for obtaining it is as follows:

[0129] D i =b t ·x t

[0130] Where t∈[1,2,3,…,T].

[0131] Step 4: Constructing an objective function based on sensitive data identification samples

[0132] The attention enhancement sample dataset obtained from the above steps is D{D1,D2,…,D}. n Assume there are c categories to be identified, and the set of their categories is C{C1,C2,…,C…}. n The set of probability matrices for all its categories is P{P1,P2,…,P}. n}

[0133] First, construct two function mappings to obtain the word feature set F{F1,F2,…,F} corresponding to each text. n The set of sentence features corresponding to each text is F′={F′1,F′2,…,F′}. n The specific calculation of F is as follows:

[0134] F = f(D)

[0135] The specific calculation of F′ is as follows:

[0136] F′=f′(D

[0137] Secondly, obtain the final set of each text feature F″={F′1′,F′2′,…,F′ n ′}.

[0138] The specific calculation of F″ is as follows:

[0139] F″=WF+W′F′

[0140] Among them, W k The weight matrix for word features is obtained by taking the frequency of each number in the weighted average of the rectangular array of the dataset. W k ′ The weight matrix is ​​constructed for sentence features.

[0141] Then, a classification function is constructed based on the final text feature set, transforming each final text feature into a probability set corresponding to each category. The calculation formula is as follows:

[0142]

[0143] The softmax function is a normalized exponential function.

[0144] Then, based on the probability set of each category and the probability matrix set of all categories to which it truly belongs, the error is calculated. The error calculation is as follows:

[0145]

[0146] Finally, by continuously adjusting the functions and parameters to reduce the loss, the optimal model is obtained. Therefore, the objective of this invention can be transformed into the following formula:

[0147]

[0148] Therefore, the function T m It serves as the loss function for subsequent deep neural network models.

[0149] Step 5: Model Construction and Training

[0150] The sensitive data identification model used in this invention is a deep neural network model. The network parameters are set as follows: 8 layers, 100 neurons in the middle layers, a learning rate of 1.5, c neurons in the final layer, a sigmoid activation function, and a loss function of T. m The gradient descent algorithm is used to make the loss function T... m Minimize, and thus obtain T m When minimized, the set of class probabilities

[0151] Then from the set To find the maximum value, use the following formula:

[0152]

[0153] At this time, P max The corresponding category is the output category of the deep neural network.

[0154] Step Six: Sensitive Data Identification

[0155] After the model is trained using training samples, the trained model is used to classify and identify unlabeled samples, and the input data is judged as sensitive data based on the output category of the deep neural network.

[0156] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0157] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution device.

[0158] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0159] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0160] In the description of this specification, references to terms such as "an embodiment," "some embodiments," "example," "specific example," or "embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0161] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

[0162] The method, apparatus, processor, and computer-readable storage medium for sensitive data identification in the context of the present invention can effectively improve the accuracy of sensitive data identification. In particular, the present invention uses an attention mechanism and constructs a specific loss function, which can comprehensively consider the concealment of sensitive data in big data and enhance attention to specific features, thereby effectively improving the robustness and generalization ability of the model.

[0163] In this specification, the invention has been described with reference to specific embodiments thereof. However, it will be apparent that various modifications and variations can be made without departing from the spirit and scope of the invention. Therefore, the specification and drawings should be considered illustrative rather than restrictive.

Claims

1. A method for identifying sensitive data in a domestically developed information technology environment, characterized in that, The method includes the following steps: (1) Collect metadata and sample data; (2) Perform data preprocessing and dataset partitioning on the collected data; (3) Construct sensitive data identification samples and perform attention enhancement operations to obtain attention-enhanced sample D; (4) Construct an objective function based on the attention-enhanced sample D, and transform the objective function into the loss function of the subsequent deep neural network model. ; (5) Using the aforementioned loss function Model building and training of deep neural network models; (6) Use the trained deep neural network model to identify and process sensitive data; Step (4) specifically includes the following steps: (4.1) Suppose that the attention-enhanced sample D contains There are 12 categories to be identified, and the set of categories to which they belong is 12. The set of probability matrices for all its categories is ; (4.2) Construct two function mappings to obtain the word feature set corresponding to each text. and the sentence feature set corresponding to each text ,in, , ; (4.3) Based on the above steps, obtain the final feature set for each text. ,in, in, The weight matrix for word features. The weight matrix constructed for sentence features; (4.4) Based on the final set of text features obtained Construct a classification function and transform the final features of each text into a probability set corresponding to each category. The calculation formula is as follows: The softmax function is a normalized exponential function. (4.5) Based on the probability set of each category and the probability matrix set of all categories to which it truly belongs, the error function loss is calculated in the following way: (4.6) By continuously adjusting various functions and parameters, the error is reduced, and the objective function is transformed into the loss function of the subsequent deep neural network model. Specifically: 。 2. The method for sensitive data identification in a domestically developed information technology environment according to claim 1, characterized in that, Step (1) specifically includes the following steps: (1.1) Collect a specified number of sample data from the business data table by traversing the data table, thereby obtaining structured and unstructured data samples and metadata information in the database and file system; (1.2) Filter and generalize the collected metadata and sample data to remove impurity data.

3. The method for sensitive data identification in a domestically developed information technology environment according to claim 2, characterized in that, Step (2) specifically includes the following steps: (2.1) For the character features in the collected relevant data, one-hot encoding is used for numerical conversion; (2.2) The MinMaxScaler() function is used to normalize the collected data so that the preprocessed data is limited to the range [0, 1]. (2.3) Divide the data into a training set and a test set, and label the training set. The training set is used to train the model, and the test set is used to verify the accuracy of the model.

4. The method for sensitive data identification in a domestically developed information technology environment according to claim 3, characterized in that, Step (3) specifically includes the following steps: (3.1) Construct a sample dataset for sensitive data identification as follows: and reference the query vector q; (3.2) Based on the sensitive data identification, the association between the sample data set X and the query vector q is obtained, and a score result is input into the softmax function for normalization mapping to obtain the attention distribution of each sample; (3.3) Finally, based on the calculated attention distribution results, attention selection is performed on the sensitive data identification sample data set X, and important information is assigned a higher weight, while unimportant information is assigned a lower weight; (3.4) Repeat the iterative process based on step (3.3) to obtain the final attention-enhanced sample D.

5. The method for sensitive data identification in a domestically developed information technology environment according to claim 4, characterized in that, Step (3.2) is specifically implemented through the following calculation formula: Where q is the query vector. The length of the vector. Let t be the t-th feature of the input sample. T is the number of features in the sample, and the scoring function is... Calculated in the following way: in, For attention distribution The transpose of the matrix, Let t be the t-th feature of the input sample. , These are all learning adjustment parameters that are manually set during the training process, and their values ​​are generally within the range of [0, 1].

6. The method for sensitive data identification in a domestically developed information technology environment according to claim 4, characterized in that, The attention-enhanced sample In D Specifically, it is obtained through the following formula: in, , .

7. The method for sensitive data identification in a domestically developed information technology environment according to claim 1, characterized in that, Step (5) specifically includes the following steps: (5.1) A deep neural network model is used for model construction; (5.2) Set the network parameters of the deep neural network model as follows: 8 layers, 100 neurons in the middle layers, a learning rate of 1.5, and a number of neurons in the final layer. The activation function is the Sigmoid function, and the loss function is... ; (5.3) The loss function is made easier by using the gradient descent algorithm. Minimize, and thus obtain When minimized, the set of class probabilities ; (5.4) The following method is used to obtain the class probability set. In the middle, obtain the maximum value: Currently, the aforementioned The corresponding category is the output category of the deep neural network model.

8. The method for identifying sensitive data in a domestically developed information technology environment according to claim 7, characterized in that, The specific steps (6) are as follows: After the deep neural network model is trained using training samples, the trained deep neural network model is used to classify and identify unlabeled samples, and the output category of the deep neural network model is used to determine whether the input data is sensitive data, so as to complete the final data identification.

9. A device for identifying sensitive data in a domestically developed information technology environment, characterized in that, The device includes: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the steps of the method for sensitive data identification under the information technology innovation environment as described in any one of claims 1 to 8.

10. A processor for identifying sensitive data in a domestically developed information technology environment, characterized in that, The processor is configured to execute computer-executable instructions, which, when executed by the processor, implement the steps of the method for sensitive data identification under the information technology innovation environment as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, It stores a computer program that can be executed by a processor to implement the various steps of the method for sensitive data identification in the information technology innovation environment as described in any one of claims 1 to 8.

Images